# How 1,000 Minds works

> A visual population lab: give a problem to sampled perspectives, watch them diverge, evaluate, and converge. One answer. Not a chatbot.

Run it at [https://1000minds.uselessai.in](https://1000minds.uselessai.in). Sign in with Google.

## What it is

You are not chatting with one model. The lab generates many perspectives for *this wording*, scores them, keeps some outliers, and synthesizes **one** answer. The UI calls those seats sampled perspectives. It does not claim independent agents.

## What to ask

Good fits: a decision, a product or ops design, “what are we missing?” Bad fits: trivia, a single coding task, a follow-up in a chat thread.

Examples you can paste:

- Clinic no-shows are 18%. Fix it without another reminder app.
- Should we take this acquisition offer this quarter, given an eight-week product bet?
- Design a weekly research brief a two-person team can actually ship.
- What are we missing if we position this as a lab, not a chatbot?
- How should a seed team pick between deepening one workflow and adding three features?

## How to ask

Name the constraint (time, team, what you refuse to build). One problem per run. Optional population (“marketers”, “clinicians”) if you want that room. Optional file if the ask lives in a brief or a screenshot. Then pick size, depth, and whether web grounding is on (it starts off). Step-by-step: [How to use the lab](/docs/using.md). What the toggles mean: [Features](/docs/features.md).

## Versus one chat reply

A chat gives you the model’s first path. This lab searches a population, ranks it, and commits to one mechanism you can read on the canvas. It is still a model. Check the answer.

## Next

- [How to use the lab](/docs/using.md)
- [Web search](/docs/web-search.md)
- [AI skill](https://github.com/shuklaji28/ai-skills): same method in Cursor, Claude, or any agent

## FAQ

### What is 1,000 Minds?

A visual population lab. You give it one problem; it generates sampled perspectives, scores them, and synthesizes one concrete answer. It does not chat turn by turn.

### Are the minds independent agents?

No. Minds are slot-assigned perspectives inside structured generation. The claim is minds per request, not secret independent agents or objective truth.

### What should I ask?

A decision, a design, or “what are we missing?” with a real constraint. Not trivia, not “write this function,” not a chat follow-up. One problem per run.

### Do I need my own API key?

No for a short hosted trial: three runs per Google account, up to 50 minds, quick depth. Save a Cerebras key for more runs, balanced depth, larger populations, or deep research on a clicked mind.

### Can I use the method without this website?

Yes. The public AI skill at https://github.com/shuklaji28/ai-skills runs the same population-search method in Cursor, Claude, or any agent. It does not call this app.